Time scales in the cloud
April 8, 2019 Leave a comment
This is still a draft of a very brief post to present and discuss time scales in the cloud.
The table below is a summary of the data needed to better understand the potential performance of distributed databases.
Let assume that 100 bytes of data is to be copied. I will assume that the network had a bandwidth of 10Gb, hence it should take approximately 100ns on the wire to send the data across.
| scope | communication | time | latency | failure |
|---|---|---|---|---|
| VM | shared memory | 100 ns | memory latency | software |
| server | virtual network | 500 ns | serialization | hardware |
| rack | physical network | 1 us | network | hardware |
| data center | physical network | 3 us | network | data center |
| different building | physical network | 20 us – 100 m | network | data center |
| remote data center | physical network | 20 ms – 1000 km | network | nuclear war |
| different continent | physical network | 120 ms – 6000 km | network | global nuclear war |
Discussion to follow
Software Transactions 